Hacker News Reader

>Codex was trained on “millions of public repositories”

if people can learn from your code, so can AIs

If people aren't allowed to copy that code into their own commercial programs, neither can AIs..

people taking short segments of code that are implementations of common algorithms and claiming those as "copied" by copilot are wrong. it didn't copy, it learned and wrote.

there's multiple examples of it outputting non-trivial code that is identical up to and including comment strings

If microsoft wants a code AI they're free to create their own training data set instead laundering copyright violation of anything that's touched github. It being "hard" isn't an excuse.

If people can learn from your face, so can cameras.

It's not so simple.

Cameras are generally legal to use to record faces though?

Publishing or selling the photo can get tricky, though, and that's why model releases exist.

That’s the difference between copying and learning though

Reply on news.ycombinator.com